1OpenAI closes the first $11 billion of its new round at a $1.25 trillion valuation
OpenAI has raised $11 billion so far in a new round valuing it at $1.25 trillion, Bloomberg reported Friday. The round has not closed: it is expected to total about $20 billion in the coming weeks - short of the $30 billion target at a $1.4 trillion valuation Bloomberg reported in late September.
In plain English
The biggest private funding round ever attempted is filling up, but smaller and cheaper than OpenAI first hoped. The $11 billion is real money already committed; the rest is still being negotiated.
Why it matters
The same day, Bloomberg reported potential Anthropic IPO buyers eyeing a valuation as low as $1.1 trillion. Both AI giants' financings are being repriced after the year's sharpest tech selloff - public-market caution is reaching private AI money. OpenAI's own IPO timeline now points to 2027, per Reuters.
SourcesTechCrunch's report·Bloomberg's original report
2Anthropic explains how Claude's memory works - dreaming included
Anthropic published a research post Friday on how it builds memory for Claude. The design borrows from the human brain: Claude tags notable events as they happen, saves them as memory notes, and can replay its activity logs while idle - "dreaming," the researchers call it - to consolidate them into long-term knowledge.
In plain English
An assistant that remembers everything remembers nothing useful. Judgment about what is worth keeping beats volume - and letting the model replay its day while idle beats fixed rules about what to write down.
Why it matters
Memory is the feature turning chatbots into assistants. Anthropic found rigid instructions ("record only failures") actually weakened its models; the lesson for anyone building agents is to guide the model's judgment, not script it.
SourcesAnthropic's research post
3DeepMind's AI agents find much faster attacks on legacy encryption
Google DeepMind researchers used supervised AI agents to discover new attacks on legacy 1024-bit RSA encryption, Chief Scientist Pushmeet Kohli announced Friday. The standout: an algorithm that, in theory, improves on the best known attack for 1024-bit keys by about 1,600 times; the agents also cut the cost of batch-factoring each additional key about 30-fold.
In plain English
AI is getting dramatically better at the math used to pick old digital locks. The locks protecting today's banking, messaging and software updates are a newer, bigger kind - 2048-bit and up - and DeepMind says the findings do not threaten them.
Why it matters
Cryptanalysis advances used to take years of manual research; agents are compressing that timeline - the same race OpenAI's GPT-6 Cyber, released Thursday, is built for. The practical message for anyone still running 1024-bit RSA: retire it.
SourcesInteresting Engineering's report·Pushmeet Kohli's announcement
Checked against original announcements and named press reports on the morning of November 7, 2026. OpenAI's funding figures are Bloomberg's reporting, citing unnamed sources; the round has not closed. DeepMind's speedup figures are its own research claims, and the 1,600x figure is a theoretical improvement against legacy 1024-bit RSA, not modern encryption. Anthropic's memory findings are its own evaluations of its own models.
